You have spent years making documents look good. Branded templates, clean PDFs, tidy Word files with headers in the right font. All of that effort was for human readers. The readers have changed.
More of your documentation is now consumed by AI before a human ever sees it. Proposals summarised by ChatGPT, SOPs fed into Claude, meeting notes parsed by an agent building your next report. The format that makes a document useful to a machine is not the format that makes it look nice in an email. And the quiet winner in this shift is Markdown.
The Format Nobody Expected to Come Back
Markdown was invented as a lightweight way to write for the web. Plain text, a few symbols for headers and lists, no visual styling baked in. For years it lived in developer circles while the rest of us wrestled with Word and Google Docs.
Then large language models arrived, and Markdown turned out to be the format they understand best. Anthropic’s own guidance for Claude notes that Markdown and XML tags help structure prompts clearly, making it easier for the model to parse your intent. The same logic applies to any document you feed in. Clean hierarchy, explicit structure, no hidden formatting noise.
If you want to understand why this matters at a deeper level, the way models interpret the structure around your words shapes the output you get back. Format is context. Context is everything.
Why This Matters for Your Token Budget
Every time you send a document to an AI, you pay in tokens. Tokens are how models measure input and output, and they have a direct cost, in money and in comprehension. A bloated document burns through the context window faster, leaving less room for the model to actually think.
A PDF exported from a designed template often carries hidden formatting, embedded fonts, and layout artefacts that translate into extra tokens when parsed. Markdown carries almost none of that. The same information, expressed in Markdown, can consume a fraction of the tokens that the same content in a Word file or PDF would. That means cheaper queries, longer effective memory, and sharper answers.
I have written before about how tokens actually work and why they shape every AI interaction. Once you see documentation through that lens, the argument for Markdown stops being a preference and starts being an economic one.
Stop Designing Documents. Start Structuring Them.
The old instinct was to make a document look professional. The new instinct should be to make it legible to a machine that cannot see. Headers that describe sections, bullet points that break up logic, short declarative sentences, explicit labels for what each section contains. No hidden meaning in bold red text or a cleverly placed table.
This is a genuine mindset shift for most business owners. You are not writing for a reader who will scan visually. You are writing for a system that will lift structure from the text itself. Every H2, every bullet, every plain sentence is a signal the model uses to work out what you meant.
For smaller businesses especially, this matters because you do not have a documentation team. You have you, a few colleagues, and an increasing number of AI tools. Making your SOPs, briefs, and internal notes Markdown-first means every one of those tools works better with less effort.
What to Do This Quarter
Start with the documents your team and your AI tools touch most often. Client briefs, onboarding notes, campaign plans, standard operating procedures. Convert them to Markdown, or at least to a structured plain-text format with clear headings and lists. Stop exporting to PDF by default.
If you are running AI agents across parts of your operation, their performance depends heavily on how you feed them information. Feed them Markdown. Watch the quality of outputs improve without changing a single prompt.
For owner-operators newer to all this, the broader primer on using AI in a small UK business is a reasonable starting point before you rip up your templates.
The takeaway is simple. The audience for your documentation has expanded to include machines, and those machines prefer plain structure to pretty design. Optimise for the reader that is now doing most of the reading. Everything else follows.
